DPM算法Windows环境

1 Introduction

This note describes some recent advances that improve the performance of the object detection

system described in [2]. Some of the improvements included here comprised the UoC-TTI LSVM-

MDPM entry in the PASCAL VOC 2009 comp3 challenge [1] and others were developed subse-

els, where each component is bilaterally symmetric. We now use a richer class of models, where

each object class is represented by a three component mixture of asymmetric models. Bilateral

asymmetry allows each component to specialize at the task of detecting left or right object poses.

During detection each component is matched to the image in both left and right orientations. This

means that in effect we have a mixture model with six components, with the extra constraint that

components are grouped into left-right symmetric pairs. The left-right pose distinction is automat-

ically learned by our system in an unsupervised fashion without the use of additional pose labels

(we ignore the incomplete pose labels given by the PASCAL annotations).

2.1 Left-Right Pose Clustering

The input data is a set of images containing instances of an object class. The location and extent of

and its flipped counterpart may be placed into the same cluster. The algorithm begins by selecting

(uniformly at random) an example and its flipped counterpart. These feature vectors seed the two

Figure 1: Fully trained car and person models using the PASCAL VOC 2007 train+val data. Here

we display 3 components for each class. In practice each component also has a left-right flipped

counterpart for a total of 6 components per class.

Figure 2: Examples of root filters trained on left-right pose clustered data from the classes (a)

bicycle, (b) horse, and (c) person.

local minima. The clustering with the best SSD objective function value is selected. Empirically

we have found that this process is very effective at clustering most object classes into two clusters

corresponding to left and right facing examples. See Figure 2 for some examples. Naturally there

are some exceptions (e.g., bottle) where the left-right distinction does not make sense.

2.2 Part Initialization

Given a root filter, we initialize k d×d part filters at twice the spatial resolution of the root by

selecting part locations and cropping out sub-arrays from a higher resolution version of the root

filter. The default configuration of the system uses k = 8 and d = 6. That is, we have 8 6×6 parts

in each deformable model.